| |||||||||||
|
Acivs 2008 Abstracts
Regular papers
Paper 109: Efficient and Flexible Cluster-and-Search for CBIR
Content-Based Image Retrieval is a challenging problem both in terms of effectiveness and efficiency. In this paper, we present a flexible cluster-and-search approach that is able to reuse any previously proposed image descriptor as long as a suitable similarity function is provided. In the clustering step, the image data set is clustered using a hybrid divisive-agglomerative hierarchical clustering technique. The obtained clusters are organized in a tree that can be traversed efficiently using the similarity function associated with the chosen image descriptors. Our experiments have shown that we can improve search-time performance by a factor of 10 or more, at the cost of small loss in effectiveness (typically less than 15%) when compared to the state-of-the-art solutions.
Paper 110: Face Recognition using Symbolic KPCA Plus Symbolic LDA in the framework of Symbolic Data Analysis: Symbolic Kernel Fisher Discriminant Method
In this paper we present a new approach called as symbolic kernel Fisher discriminant analysis (symbolic KFD) for face recognition based on symbolic kernel principal component analysis (symbolic KPCA) and symbolic linear discriminant analysis (symbolic LDA) in the framework of symbolic data analysis. It is well known that the distribution of face images, under a perceivable variation in view point, illumination and facial expression is highly nonlinear and complex. The linear techniques based on symbolic LDA cannot provide reliable and robust solutions to such face recognition problems because these techniques fail to capture a non-linear relationship with linear mapping. However, proposed symbolic KFD method overcomes this limitation by using kernel trick to represent complicated nonlinear relations of input data. The classical KFD method uses single valued variables to represent the facial features, where as, the proposed symbolic KFD extract interval type non linear discriminating features, which are robust due to varying facial expression, view point and illumination. The new algorithm has been successfully tested using three databases, namely, ORL database, Yale Face database and Yale Face database B. The experimental results show that symbolic KFD outperforms other KFD algorithms.
Paper 111: Motion Recovery for Uncalibrated Turntable Sequences Using Silhouettes and a Single Point
This paper addresses the problem of self-calibration and motion recovery for turntable sequences. Previous works exploited silhouette correspondences induced by epipolar tangencies to estimate the image invariants under turntable motion and recover the epipolar geometry. These approaches, however, require the camera intrinsics in order to obtain an Euclidean motion, and a dense sequence is required to provide a precise initialization of the image invariants. This paper proposes a novel approach to estimate the camera intrinsics, the image invariants and the rotation angles from a sparse turntable sequence. The silhouettes and a single point correspondence are extracted from the image sequence. The point traces out a conic in the sequence, from which the fixed entities (i.e., the image of the rotation axis, the horizon, the vanishing point of the coordinates, the circular points and a scalar) can be recovered given a simple initialization of the camera intrinsic matrix. The rotation angles are then recovered by estimating the epipoles that minimize the transfer errors of the outer epipolar tangents to the silhouettes for each pair of images. The camera intrinsics can be further refined by the above optimization. Based on a given range of the initial focal length, a robust method is proposed to give the best estimate of the camera intrinsics, the image invariants, the full camera positions and orientations, and hence a Euclidean reconstruction. Experimental results demonstrate the simplicity of this approach and the accuracy in the estimated motion and reconstruction.
Paper 112: An evaluation methodology for image mosaicing algorithms
Several image mosaicing algorithms claiming to advance the state of the art have been proposed so far. Though sometimes improvements can be recognised without quantitative evidences, the importance of a principled methodology to compare different algorithms is essential as this discipline evolves. Which is the best? What means the best? How to ascertain the supremacy? To answer such questions, in this paper we propose an evaluation methodology including standard data sets, ground-truth information and performance metrics. We also compare three variants of a well-known mosaicing algorithm according to the proposed methodology.
Paper 113: High-speed acquisition and pre-processing of polarimetric image sequences
We have developed an imaging polarimeter working in the visible range. It is composed of a CCD camera with a bistable ferroelectric liquid crystal modulator as a polarizing rotator; therefore the degree of polarization (DOP) is evaluated from two successive acquisitions. Apparent motion in the scene induces a false evaluation of the DOP, especially on the edges of the objects. We present our acquisition system and two methods to correct defects in DOP images of moving objects. First, we propose a post-processing temporal median filtering, correcting the DOP once computed. The second method consists in performing a motion estimation to correct the object's displacement between the two polarization state pictures. Comparison based on qualitative and quantitative results with real data is provided.
Paper 115: Fusion of Multi-View Tissue Classification Based on Wound 3D Model
Region classification from a single image is no more reliable when the labeling must be applied on a 3D surface. Depending on camera viewpoint and surface curvature, lighting variations and perspective effects alter colorimetric analysis and area measurements. This problem can be overcome if a 3D model of the object of interest is available. This general approach has been evaluated for the design of a complete wound assessment tool using a simple free handled digital camera. Clinical tests demonstrate that multi view classification results in enhanced tissue labeling and more precise measurements, a significant step toward accurate monitoring of the healing process.
Paper 117: An Iterative Kalman Filter Approach to Camera Calibration
An iterative camera calibration approach is presented in this paper. This approach allows computing the optimal camera parameters for a given set of data. If non linear estimation process is done, a risk of reaching a local minimum exists. With this method this risk is reduced and a best estimation is achieved. By one hand, an iterative improving of the estimated camera parameters is done maximizing a posteriori probability density function (PDF) for a given set of data. To resolve it, a Kalman filter is used based on the Bayesian standpoint. Each update is carried out starting with a new set of data, its covariance matrix and a previous estimation of the parameters. In this case, a different management of the input data is done to extract all its information. By the other hand, apart from the calibration algorithm, a method to compute an interval which contains camera parameters is presented. It is based on computing the covariance matrix of the estimated camera parameters.
Paper 118: Face Recognition Using Parabola Edge Map
Several applications such as access control, behaviour observation and videoconferencing require a real-time method for face recognition. We propose to represent faces with parabola segments with an algorithm that allows us to fit parabola segments real-time to edge pixels. Parabola offer a good description for the many edges in a face, which is advantageous for several applications such as recognition of the facial expression and facial orientation. We use parabola segments for face recognition, which is done by a technique that matches parabola segments from different faces, based on distance and intensity.
Paper 119: A new algorithm for dominant point detection by quasi-collinear break points supression
A new algorithm for dominant points of digital planar curves is presented. This algorithm is relied on a new method for quasi-collinear point supression. All the break points are obtained from the initial contour. The proposed algorithm deletes quasi-collinear break points using a variable distance as treshold value. The resultant points are dominant points. This aproximation is compared with other algorithms. The experimental results show that this method works well for digital planar curves with features of several sizes.
Paper 122: Robust curvature extrema detection based on new numerical derivation
Extrema of curvature are useful key points for different image analysis tasks. Indeed, polygonal approximation or arc decomposition methods used often these points to initialize or to improve their algorithms. Several shape-based image retrieval methods focus also their descriptors on key points. This paper is focused on the detection of extrema of curvature points for a raster-to-vector-conversion framework. We propose an original adaptation of an approach used into nonlinear control for fault- diagnosis and fault-tolerant control based on algebraic derivation and which is robust to noise. The experimental results are promising and show the robustness of the approach when the contours are bathed into a high level speckled noise.
Paper 124: Spherical edge detector : application to omnidirectional imaging
In this paper, we present an efficient approach to detect edges in omnidirectional images. The main problem with such images, is that resolution is very high in the periphery and poor in the center. Applying the classical (planar) operators to these images will introduce errors. We propose to map the image on the unit sphere of equivalence, and to construct edge and smooth operators in this new space. The effect of the proposed edge detector is the same over the image. We will show that results obtained by our image processing tools give better results than classical edge detection operators.
Paper 128: Network Security Using Biometric And Cryptography
We propose a biometrics-based(fingerprint)Encryption/Decryption Scheme, in which unique key is generated using partial portion of combined sender's and receiver's fingerprints. From this unique key a random sequence is generated, which is used as an asymmetric key for both Encryption and Decryption. Above unique Key is send by the sender after Watermaking it in sender's fingerprint along with Encrypted Message. The computational requirement and network security features are addressed. Proposed system has a advantage that for public key, it has not to search from a database and security is maintained. This paper reports on work in progress.
Paper 129: Human Pose Estimation in Vision Networks via Distributed Local Processing and Nonparametric Belief Propagation
In this paper we propose a self-initialized method for human pose estimation from multiple cameras. A graphical model for the articulated body is defined through explicit kinematic and structural constraints, which allows for any plausible body configuration and avoids learning the joint distributions from training data. Nonparametric belief propagation (NBP) is used to infer the marginal distributions. However, to address the problem of the inference being trapped in local optima and to achieve fast convergence, a reasonably good pose initialization is required. A bottom-up approach is used to detect body parts distributedly in local processing of each camera. 3D Geometry correspondence relates 2D camera observations spatially to generate a rough pose estimation to initialize node marginal distribution. The marginal distributions are then refined through NBP. Estimated 3D body joint positions are quantitatively analyzed with motion capture data.
Paper 131: An efficient hardware architecture without line memories for morphological image processing
In this paper, we present a novel hardware architecture to achieve erosion and dilation with a large structuring element. We are proposing a modification of HGW algorithm with a block mirroring scheme to ease the propagation and memory access and to minimize memory consumption. It allows to suppress the needs for backward scanning and gives the possibility for hardware architecture to process very large lines with a low latency. It compares well with the Lemonnier's architecture in terms of ASIC gates area and shows the interest of our solution by dividing the circuit area by an average of 10.
Paper 133: A geometric primitive extraction process for remote sensing problems
This paper presents a new approach to describe images in terms of geometric objects. Methods based on conventional stochastic marked point processes have already led to convincing image analysis results but possess several drawbacks such as complex parameter tuning, large computing time, and lack of generality. We propose a generalized marked point process model which can be performed in shorter computing times and applied to a variety of applications without modifying the model or tuning parameters. In our approach, both linear and areal primitives extracted from a library of geometric objects are matched to a given image using a probabilistic Gibbs model. A Jump-Diffusion process is performed to find the optimal object configuration. Experiments with remotely sensed images show good potentialities of the proposed approach.
Paper 135: Geodesic Active Contours with Combined Shape and Appearance Priors
We present a new object segmentation method that is based on geodesic active contours with combined shape and appearance priors. It is known that using shape priors can significantly improve object segmentation in cluttered scenes and occlusions. Within this context, we add a new prior, based on the appearance of the object, (i.e., an image) to be segmented. This method enables the appearance pattern to be incorporated within the geodesic active contour framework with shape priors, seeking for the object whose boundaries lie on high image gradients and that best fits the shape and appearance of a reference model. The output contour results from minimizing an energy functional built of these three main terms. We show that appearance is a powerful term that distinguishes between objects with similar shapes and capable of successfully segment an object in a very cluttered environment where standard active contours (even those with shape priors) tend to fail.
Paper 136: Effective Segmentation for Dental X-ray Images Using Texture-Based Fuzzy Inference System
In teeth-related radiograph research, the information of teeth shape is the most critical factor for achieving highly automated diagnosis. Therefore, accurate segmentation is an essential but difficult task due to low contrast and uneven exposure of the dental X-ray image. In this paper, we propose a novel scheme to automatically segment teeth by using texture characteristics instead of primitive intensity or edge used in previous researches. At first, image enhancement based on homogeneity measurement is applied to accentuate the texture of gums while smoothing the teeth so that a coarse clustering result can be obtained. Then, fuzzy inference is applied to speculate degrees of pixel belonging to either part. Finally, region growing based on inferences is performed to obtain the complete shape of teeth. The experimental results show that our proposed method indeed outperforms the methods using direct intensity or edge in segmenting complete teeth from X-ray dental images.
Paper 137: Nonparametric level-set segmentation based on the minimization of the stochastic complexity
In this paper, a novel non parametric method of image segmentation is deduced from the stochastic complexity principle. The main advantage of this approach is that it does not rely on any assumption on the probability density functions in each region and does not include any free parameter that has to be adjusted by the user in the optimized criterion. This results in a very flexible and robust segmentation algorithm. Various simulations performed with both synthetic and real images show that the proposed non parametric algorithm performs similarly to the parametric counterparts with the flexibility of a nonparametric approach.
Paper 138: Model-Based Gait Recognition Using Multiple Feature Detection
This paper presents a gait recognition algorithm for human identification from a sequence of segmented noisy silhouettes in a low-resolution video. The main contribution of the proposed work is the use of the hierarchical recovery of a static body and stride parameters of model subjects to the walking pose. The proposed algorithm overcomes drawbacks of existing works by extracting a set of relative model parameters instead of directly analyzing the gait pattern. The feature extraction function in the proposed algorithm consists of motion detection, object region detection, and active shape model (ASM), which alleviate problem in the baseline algorithm such as; background generation, shadow removal, and higher recognition rate. Performance of the proposed algorithm has been evaluated by using the HumanID Gait Challenge data set, which is the largest gait benchmarking data set with 122 objects with different realistic parameters including viewpoint, shoe, surface, carrying condition, and time.
Paper 139: Pedestrian Detection and Tracking Based on Far Infrared Visual Information
This article discusses a pedestrian detector for an experimental vehicle, based on visual information from a single far infrared camera. The system considers three consecutive processes for each image. Once the candidates' heads have been extracted, regions of interest are resized based on the distance to the camera and then filtered by its vertical edges symmetry. Once the bounding boxes of possible pedestrians are picked a spatial correlation with some template models takes place. Finally, detected pedestrians are tracked, contrasting their position on successive frames. The results are satisfactory, classifying correctly almost 96% of pedestrians closer than 45m to the vehicle.
Paper 141: A New Segmentation Approach for Ear Recognition
Personal identification based on the ear structure is a new biometrics. Clearly ear segmentation plays a vital role in automated ear recognition systems. In this paper, a new segmentation method for ear recognition is proposed. We apply the Canny edge detector to an ear image. Then the longest path in the edge image is extracted and selected as the outer boundary of the ear. By selecting the top, bottom, and left points of the detected boundary, we form a triangle with the selected points as its vertices. Further we calculate the barycenter of the triangle and select it as a reference point in all images. Then the ear region is extracted from the entire image using a predefined window centered at the reference point. Experimental results show the effectiveness of our proposed segmentation method.
Paper 144: Use of local surface curvature estimation for adaptive vision system based on active light projection
In this paper, we present a new 3D reconstruction approach based on local surface curvature analysis. Its integration can make a normal active stereoscopic system intelligent, and capable to produce directly optimized 3D model. The iterative 3D reconstruction process begins with a sparse and regular point pattern. Based on the reconstructed 3D point cloud, the local surface curvature around each 3D point is estimated. Those 3D points located in flat areas are removed from the 3D model, and a new pattern is created to project more points onto the object where there is high surface curvature. The 3D model is thus refined progressively during the acquisition process, and finally an optimized 3D model is obtained. Our numerous experiments showed that compared to the 3D models generated by commercial system, the loss of morphological quality is negligible, and the gain by the simplification of the model is considerable.
Paper 145: Video Object Segmentation based on Feedback Schemes guided by a Low-level Scene Ontology
This paper presents a knowledge-based framework for video analysis which systematically exploits relationshipd among analysis stages. A set of step-by-step feedback paths controls feedback generation and reception between consecutive analysis stages. An analysis ontology, which includes occurrences in the scene from high to very low semantic level, controls iterative decisions on every stage. As a result, both overall and intermediate analysis results are improved. This paper presents the framework and focuses on its application to foreground objects extraction. Experimental results show that the framework provides a richer low-level representation of the scene and improved short-term change detection and foreground detection masks.
Paper 149: Robust Lip Contours Localization and Tracking Using Multi Features-Statistical Shape Models
We propose and evaluate methods for enhancing performances of lip contours localization and tracking, which are based on the concepts of Statistical Shape Models (e.g. Active Shape Models, Hybrid Active Shape Models) and optimization of multi features. A single feature-based ASM gets good perform-ance only in particular conditions but gets stuck in local minimum or gives bad performance in noisy conditions. In this paper, we propose to use 3 features: Normal Profile, Grey Level Patches and Gabor Wavelets and combine them by using a voting approach to derive a robust method (MF-ASM) on lip contours detection. Since the original ASM does not take into account the temporal information from previous frames, the lip contours are tracked by replacing the standard ASM with our hybrid ASM which is capable to take advantage of temporal information. Initial experimental results using popular audio-visual database show that our methods are more robust to the local minimum problem and give higher accuracy than traditional single feature-based ASM in lip con-tours detection and tracking.
Paper 150: Towards Fully Automatic Image Segmentation Evaluation
Spatial region (image) segmentation is a fundamental step for many computer vision applications. Although many methods have been proposed, less work has been done in developing suitable evaluation methodologies for comparing different approaches. The main problem of general purpose segmentation evaluation is the dilemma between objectivity and generality. Recently, figure ground segmentation evaluation has been proposed to solve this problem by defining an unambiguous ground truth using the most salient foreground object. Although the annotation of a single foreground object is less complex than the annotation of all regions within an image, it is still quite time consuming, especially for videos. A novel framework incorporating background subtraction for automatic ground truth generation and different foreground evaluation measures is proposed, that allows to effectively and efficiently evaluate the performance of image segmentation approaches. The experiments show that the objective measures are comparable to the subjective assessment and that there is only a slight difference between manually annotated and automatically generated ground truth.
Paper 152: Atmospheric turbulence restoration by diffeomorphic image registration and blind deconvolution
A novel approach is presented in this paper to improve images which are altered by atmospheric turbulence. Two new algorithms are presented based on two combinations of a blind deconvolution block, an elastic registration block and a temporal filter block. The algorithms are tested on real images acquired in the desert in New Mexico by the NATO RTG40 group.
Paper 153: Image Denoising using Similarities in the Time-Scale Plane
This paper presents a de-noising method that recognizes similarities in the image through the time scale behaviour of wavelet coefficients. Wavelet details are represented as linear combination of predefined atoms whose center of mass traces trajectories in the time scale plane (from fine to coarse scale). These trajectories are the solution of a proper ordinary differential equation and characterize atoms corresponding to groups of not isolated singularities in the signal. The distances among atoms, the ratio of their amplitudes and the difference of their decay along scales are the parameters to use for defining similarities in the image. Experimental results show the potentialities of the method in terms of visual quality and mean square error, reaching the most powerful and recent de-noising schemes.
Paper 155: 3D Face Recognition Evaluation on Expressive Faces Using the IV2 Database
The purpose of this paper is to study the influence of face expressions on the performance of a 3D face recognition algorithm. Three facial surface matching based algorithms, namely ICP, Localized ICP (L-ICP) and Region-based ICP (R-ICP), are benchmarked on several sets of data : the two first sets with neutral faces and the last with expressive ones. Results show that the R-ICP algorithm provides more robustness to face expression verification than the two other approaches.
Paper 157: Crowd Behavior Recognition for Video Surveillance
Crowd behavior recognition is becoming an important research topic in video surveillance for public places. In this paper, we first discuss the crowd feature selection and extraction and propose a multiple-frame feature point detection and tracking based on the KLT tracker. We state that behavior modelling of crowd is usually coarse compared to that for individuals. Instead of developing general crowd behavior models, we propose to model crowd events for specific end-user scenarios. As a result, a same type of event may be modelled slightly differently from one scenario to another and several models are to be defined. Consequently, fast modelling is required and this is enabled by the use of an extended Scenario Recognition Engine (SRE) in our approach. Crowd event models are defined; particularly, composite events accommodating evidence accumulation allow to increase detection reliability. Tests have been conducted on real surveillance video sequences containing crowd scenes. The crowd tracking algorithm proves to be robust and gives reliable crowd motion vectors. The crowd event detection on real sequences gives reliable results of a few common crowd behaviors by simple dedicated models.
Paper 160: Embedding of a real time image stabilization algorithm on SoPC platform, a chip multi-processor approach
Highly regular multi-processor architecture are suitable for inherently highly parallelizable applications such as most of the image processing domain. System on a programmable chip (SoPC) allows hardware designers to tailor every aspects of the architecture in order to match the specific application needs. These platforms are now large enough to embed an increasing number of core, allowing implementation of a multi-processor architecture with an embedded communication network.
In this paper we present the parallelization and the embedding of a real time image stabilization algorithm on SoPC platform. Our overall hardware implementation method is based upon meeting algorithm processing power requirement and communication needs with refinement of a generic parallel architecture model. Actual implementation is done by the choice and parameterization of readily available reconfigurable hardware modules and customizable commercially available IPs.We present both software and hardware implementation with performance results on a Xilinx SoPC target.
Paper 161: Nighttime Vehicle Detection for Intelligent Headlight Control
A good visibility of the road ahead is a major issue for safe nighttime driving. However, high beams are sparsely used because drivers are afraid of dazzling others. Thus, the intelligent automatic control of vehicles' headlight is of great relevance. It requires the detection of oncoming and preceding vehicles up to such a distance that only camera based approaches are reliable. At nighttime, detecting vehicles using a camera requires to identify their head or tail lights. The main challenge of this approach is to distinguish these lights from reflections due to infrastructure elements. In this paper we confront such a challenge by using a novel image sensor also suitable for other driver assistance applications. Different appearance features obtained from that sensor are used as input to a novel classifier-based module which, for each detected target, yields a degree of resemblance to a vehicle light. This resemblance is integrated in time using a novel temporal coherence analysis which allows to react in one single frame for targets that are clear vehicle lights, or in only a few frames for those whose type is more difficult to discern.
Paper 163: Semantic Map Generation from Satellite Images for Humanitarian Scenarios Applications
This paper demonstrates how knowledge driven methods and the associated data analysis algorithms are changing the paradigms of user-data interactions, providing an easier and wider access to the Earth Observation data. Some information theory based algorithms are proposed for anomaly and change detection on SPOT images, relative to a widespread humanitarian crisis scenario: floods. The outcomes of these algorithms define an informational representation of the image, revealing the spatial distribution of a particular theme. Using image analysis and interpretation, the multitude of features from a scene are classified into meaningful classes to create sematic maps.
Paper 164: Performance analysis of generalized zerotree coders varying the maximum zerotree degree
Despite the release of the JPEG-2000 standard, wavelet-based zerotree coders keep being object of intense research because of their conceptual simplicity and excellent performance. Recently, it has been shown that all zerotree coders can be described by specifying the involved data structures (typically, degree-k zerotrees) and a very limited set of tree decomposition rules, leading to the class of generalized zerotree coders. Thanks to this abstraction, defining and implementing new coders of this class becomes straightforward. In this work, we then investigate, by means of numerical experiments on various types of visual sources, the performance achievable by such coders as a function of the degree of the underlying zerotrees.
Paper 166: Fast Saliency-Based Motion Segmentation Algorithm for an Active Vision System
In this work, we propose a saliency-based approach for estimating and segmenting 3D motions of multiple moving objects represented by 2D motion vector fields (MVF). In order to overcome typical problems in autonomous mobile robotic vision such as noise, occlusions, and inhibition of the ego- motion defects of a moving camera head, a classification module has been implemented to define the global motion of the mounted camera. The proposed method achieves valuable reduction in computational time by applying a guided control module which limits the segmentation output to a flexible predefined threshold value. The computational enhancement is very important since the output of the motion segmentation approach is implemented in an active vision system.
Paper 169: Sequential Blind PSF Estimation and Restoration of Aerial Multispectral Images
Blind restoration of aerial multipspectral images, through a sequential deconvolution scheme is addressed in this paper. The proposed scheme is composed of three successive optimized processes: image denoising followed by Point Spread Function (PSF) estimation and finally image restoration. First, an iterative denoising filter is applied and stopped at the iteration when an optimal estimation of the blurry image is obtained. Secondly, slighty TV (Total Variation) regularized PSF estimation is carried out on an almost noise free version of the blurry image. In order to keep unknown original image fixed during PSF estimation, shocked filtering of filtered image is efficiently considered. Thirdly, assuming the previously estimated PSF fixed, a TV-regularized deconvolution is performed on filtered image to give better estimation of original image. Regularization parameters are automatically tuned using regularization-scale relationship. Results obtained on aerial CASI and AISA Eagle multispectral images prove efficacy of the proposed scheme.
Paper 170: Phase Unwrapping in Fringe Projection Systems Using Epipolar Geometry
A new method for phase unwrapping is introduced which realizes the unwrapping of phase images without binary codes produced by fringe projection systems using at least two cameras and one projector. The novelty of the method is the use of the epipolar geometry between the two cameras and the projector in order to achieve a unique point correspondence. The method is suited for systems which should realize a short recording time for the image sequence acquisition. It is very robust even at positions with abrupt change of depth.
Paper 172: An Effective Salience-Based Algorithm for Shape Matching
This paper shows the shape retrieval descriptor based on saliences by exploiting the relation between a contour and its skeleton. The saliences are a very important way to represent a shape, because they are invariants under linear transformations. We introduce a new matching algorithm to estimate the similarity between two shapes represented by its contour saliences, even when we are dealing with two shapes of the same type but with different numbers of saliences. Some experimental results are presented and discussed in order to demonstrate the potentiality of the proposed technique.
Paper 173: Reliable Eyelid Localization for Iris Recognition
This article presents a new eyelid localization algorithm based on a parabolic curve fitting. To deal with eyelashes, low contrast or false detection due to iris texture, we propose a two steps algorithm. First, possible edge candidates are selected by applying an edge detection on a restricted area inside the iris. Then, a gradient maximisation is applied along every parabola, on a larger area, to refine the parameters and select the best one. Experiments have been conducted on the CASIA-IrisV3-Interval database that have been manually segmented. A new performance measure is proposed, carried out by comparing the segmented images obtained by the proposed method with the manual segmentation.
Paper 174: A Robust Method for Filling Holes in 3D Meshes based on Image Restoration
In this work a method for filling holes in 3D meshes based on a 2D image restoration algorithm is expounded. Since 3D data must be converted to a suitable input format, a 3D to 2D transformation is executed by projecting the 3D surface onto a grid. The storage of the depth information in every grid provides the 2D image which the restoration algorithms is applied in. Finally, an inverse transformation 2D to 3D is performed and the new produced data added to the damaged mesh. To test the method, artificial holes have been generated on a set of 3D surfaces. The distances between 3D original surfaces (before damaging it) and 3D repaired ones have been measured and a comparison with a commercial software has been established. Furthermore, the relation between holes areas and success rates has been also studied. This method has been applied to the sculptures of the collection from the National Museum of Roman Art in Spain with good results.
Paper 176: "Local Rank Differences" Image Feature Implemented on GPU
A currently popular trend in object detection and pattern recognition is usage of statistical classifiers, namely AdaBoost and its modifications. The speed performance of these classifiers largely depends on the low level image features they are using: both on the amount of information the feature provides and the executional time of its evaluation. Local Rank Differences is an image feature that is alternative to commonly used haar wavelets. It is suitable for implementation in programmable (FPGA) or specialized (ASIC) hardware, but – as this paper shows – it performs very well on graphics hardware (GPU) as well. The paper discusses the LRD features and their properties, describes an experimental implementation of LRD in graphics hardware, presents its empirical performance measures compared to alternative approaches and suggests several notes on practical usage of LRD and proposes directions for future work.
Paper 178: Head Pose Determination using Synthetic Images
In this paper, we propose a new approach to determine the head pose which is a very important issue in several new applications. Our method consists in building a synthetic image database for a dense set of pose parameter values. This can be done with only one real image of the face using the Candide-3 model. To determine the pose, we compare each synthesized face image to the current image using a specific metric. This metric is an Hausdorff distance like applied to gradient orientation features. Experimental results shows the efficiency of our approach on real images. The improvement is also proved through a comparison with other technique presented in literature.
Paper 181: A Real-Time Vision System for Traffic Signs Recognition Invariant to Translation, Rotation and Scale
In this paper a system is presented for real time recognition of the traffic signs. Sign detection is done by a method of adaptively growing window. Classification is based on matching of the modulo shifted phase histograms. These are built from the stick component of the structural tensor rather than from an edge detector. To cope with inherent rotations of signs a novel measure is proposed for matching of the modulo shifted histograms that also boosts responses of highly probable values. The method is tolerant of small translations, rotations and symmetrical changes of scale. It works also well under different lighting conditions and tolerates noise and small occlusions.
Paper 182: Dynamic Selection of Characteristics for Feature Based Image Sequence Stabilization
In this document, we deal with the problem of video cameras operating outdoors where the atmospheric conditions and the vibration caused by vehicles passing by make it difficult to sustain the assumption of a fixed camera. Under these circumstances, a method like ours, that dynamically chooses the features that support the transformation from frame to frame, is different to others that fix the features that are tracked. That is because, for vehicular traffic monitoring applications, there are features that appear or disappear as time passes by. This contributes to the flexibility and capability of adaptation of the method to different scenarios. To test the algorithms, we used several image sequences of a camera, experiencing heavy motion due to wind and/or vibration due to vehicular traffic.
Paper 183: Real-time wavelet-spatial-activity-based adaptive video enhancement algorithm for fpga
In this paper we present a new wavelet-based video enhancement algorithm, designed for highly optimized dedicated ICs. The new algorithm is verified on FPGA platform with target being real-time video processing. The main application of the proposed scheme is a high definition (HD) TV, where we consider visibly annoying video coding artifacts and noise (assumed as white Gaussian).
In the proposed denoising scheme each video frame is processed independently, i.e., only spatial filtering is performed. Specifically, two-dimensional (2D) non-decimated wavelet transform is applied to the frame, after which the proposed feature and noise adaptive shrinkage operation on the wavelet coefficients is done. Finally, the denoised image is reconstructed by inverse wavelet transform. The main contribution of the paper is the proposed (i) hardware-friendly scheme for the wavelet decomposition - reconstruction framework with full parallelism and reduced memory resources required and (ii) efficient and low computationally expensive feature adaptive shrinkage algorithm for denoising.
The designed framework is verified in SystemC and on FPGA Gold Chipit platform with WXGA Panel. The annoying artifacts and noise are shown to be efficiently removed with small or no visible reduction in spatial resolution.
Paper 190: A Multicomponent Image Segmentation Framework
In this paper, we propose a framework for the segmentation of multicomponent images. The specific framework we aim at contains different steps in which all components of the multicomponent image are processed simultaneously, accounting for the correlation between the image components. The framework contains the following steps:
a) to initiate, a pixel-based, spectral clustering procedure is applied.
b) to include spatial information, a model-based region-merging technique is used, applying a multinormal model for the coefficient regions, and estimating the model parameters using Maximum Likelihood principles;
c)the model allows to treat noise that might be present efficiently;
d) a multiscale version of the framework is established by repeating the same procedure at different resolution levels of the original image.
e) Then, a link between the different levels is established by constructing a hierarchy between the regions at different levels.
In this work, we will demonstrate the performance of the framework for segmentation purposes. The procedure is performed on color images and multispectral remote sensing images.
Paper 191: Adaptive metadata management system for distributed video content analysis
Scientific advances in the development of video processing algorithms now allow various distributed and collaborative vision-based applications. However, the lack of recognized standard in this area drives system developers to build specific systems, preventing from e.g. content analysis components upgrade or system reuse in different environments. As a result, the need for a generic, context-independent and adaptive system for storing and managing video analysis results comes out as conspicuous. In order to address this issue, we propose a data schema independent data warehouse backed by a multi-agent system. This system relies on the semantic web knowledge representation format, namely the RDF, to guarantee maximum adaptability and flexibility regarding schema transformation and knowledge retrieval. The storage system itself, namely data warehouse, comes from the state-of-the-art technologies of knowledge management, providing efficient analysis and reporting capabilities within the monitoring system.
Paper 192: An Interval-valued Fuzzy Morphological Model based on Lukasiewicz-Operators
Mathematical morphology is a well-known theory to process binary, grayscale or color images. In this paper, we introduce interval-valued fuzzy mathematical morphology as an extension of classical and fuzzy morphology. It originates from the observation that the pixel values of a grayscale image are not always certain, and models this uncertainty using interval-valued fuzzy set theory. In this way, we are able to incorporate the uncertainty regarding measured pixel values into the toolbox of morphological operators. We focus our attention on a morphological model whose underlying logical framework is based on the Lukasiewicz-operators. For this model we investigate and discuss general theoretical properties, some computational aspects, as well as its relation to fuzzy morphology and classical grayscale morphology.
Paper 193: Decision Trees in Binary Tomography for Supporting the Reconstruction of hv-Convex Connected Images
In binary tomography, several algorithms are known for reconstructing binary images having some geometrical properties from their projections. In order to choose the appropriate reconstruction algorithm it is necessary to have a priori information of the image to be reconstructed. In this way we can improve the speed and reduce the ambiguity of the reconstruction. Our work is concerned with the problem of retrieving geometrical information from the projections themselves. We investigate whether it is possible to determine geometric features of binary images if only their projections are known. Most of the reconstruction algorithms based on geometrical information suppose hv-convexity or connectedness about the image to be reconstrcuted. We investigate those properties in detail, and also the task of separating 4- and 8-connected images. We suggest decision trees for the classification, and show some preliminary experimental results of applying them for the class of hv-convex and connected discrete sets.
Paper 194: Curvature Estimation and Curve Inference with Tensor Voting: a New Approach
Recently the tensor voting framework (TVF), proposed by Medioni at al., has proved its effectiveness in perceptual grouping of arbitrary dimensional data. In the computer vision field, this algorithm has been applied to solve various problems as stereo-matching, boundary inference, and image inpainting. In the last decade the TVF was augmented with new saliency features, like curvature and first order tensors.
In this paper a new curvature estimation technique is described and its effectiveness, when used with the saliency functions proposed in [1], is demonstrated. Results are shown for synthetic datasets in spaces of different dimensionalities.
Paper 199: A New Ground Movement Compensation Approach for Obstacle Detection Using an In-Vehicle Camera
The purpose of this paper is to propose a new approach to detect obstacles using a single camera mounted on a vehicle when the vehicle is backing or turning round at an intersection at a low speed. Using equations among feature point locations and optical flows in geometrically transferred top view images, ground movement information can be estimated. We compensate for the ground movement between consecutive top view images using the estimated ground movement information and compute the difference image using the compensated previous top-view image and current top-view image. Finally, obstacle regions in top-view images can be easily extracted using the difference image. The proposed approach is demonstrated to be effective by evaluation images captured by an in vehicle camera.
Paper 200: Multidimensional noise removal based on fourth order cumulants
This paper presents a new multi-way filtering method for multi-way images impaired by correlated Gaussian noise. Instead of matrices or vectors, multidimensional images are considered as multi-way arrays also called tensors. Some noise removal techniques consist in vectorizing or matricizing multi-way data. That could lead to the loss of inter-bands relations. The presented filtering method consider multidimensional data as whole entities. Such a method is based on multilinear algebra. Most of multidimensional noise removal techniques are based on second order statistics and are only efficient in the case of additive white noise. But in some cases, it can be interesting to consider additive correlated noise. Therefore, we introduce higher order statistics for tensor filtering to remove Gaussian components. Experiments on HYDICE hyperspectral images are presented to show the improvement using higher order statistics.
Paper 202: Attention-based Segmentation on an Image Pyramid Sequence
This paper proposes a computational model of attention-based segmentation in which a sequence of image pyramids of early visual features is computed for a video sequence and a repetition of selective attention and figure-ground segmentation is performed on the sequence for object perception through successive segment development with mergence of concurrent segments. Attention is stochastically selected on a multi-level saliency map that is called a visual attention pyramid and segmentation is performed on Markov random fields which are dynamically formed around foci of attention. A set of segments and their spatial relation are stored in a visual working memory and maintained through the repetitive attention and segmentation process. Performances of the model are evaluated for basic functions of the vision system such as visual pop-out, figure-ground reversal and perceptual organization and also for real-world scenes which contain objects designed to attract attention.
Paper 208: Bit Domain Encryption
In recent years we have seen significant growth in multimedia based Internet applications and multimedia commerce. The advancements in multimedia commerce resulted has in a rapid growth, in the amount of multimedia data transferred over network. The multimedia data stored on the web servers as well as the data transferred over the network needs to be protected from piracy and eavesdropping. Hence, there is a strong need for encrypting the multimedia content. But owing to the size of the multimedia content and real time requirements for encoding, transmitting and decoding the multimedia content, usage of standard encryption/decryption schemes prove to be an overhead. Hence lightweight encryption schemes are gaining popularity. The schemes are designed using structure of the multimedia content, and partially encrypting the content such that it results in insertion of sufficient noise to make the content unintelligible. In this paper we present a scalable and secure lightweight encryption scheme for image and video data in bit domain. The scheme in addition to being secure and scalable has negligible impact on compression.
Paper 209: Foliage Recognition based on Local Edge Information
In many real-world object recognition applications, texture plays a very important role. Much research has gone into texture-based segmentation methods, which focus on finding the boundaries between uniformly textured regions. These methods can be adapted to recognize a specific type of texture. However, many naturally occuring objects have a texture with a high degree of irregularity that complicates their recognition, causing generic algorithms to fail. This paper presents a method to recognize one specific type of natural texture: foliage. Starting from a proven technique to extract texture feature information, a recognition model is constructed that incorporates prior knowledge about this particular texture. The algorithm recognizes 95% of the foliage areas in real-world video data, with less than 8% false positives.
Paper 210: A Pseudo-Logarithmic Image Processing Framework for Edge Detection
The paper presents a new [pseudo-] Logarithmic Model for Image Processing (LIP), which allows the computation of gray-level addition, substraction and multiplication with scalars within a fixed gray-level range [0; D] without the use of clipping. The implementation of Laplacian edge detection techniques under the proposed model yields superior performance in biomedical applications as compared with the classical operations (performed either as real axis operations, either as classical LIP models).
Paper 212: On bin configuration of shape context descriptors in human silhouette classification
Shape context descriptors have been a valuable tool in shape description since their introduction. In this paper we examine the performance of shape context descriptors in the presence of noisy human silhouette data. Shape context descriptors have been shown to be robust to Gaussian noise in the task of shape matching. We implement four different configurations of shape context by altering the spacing of the histogram bins and then test the performance of these configurations in the presence of noise. The task used for these tests is recognition of body part shapes in human silhouettes. The noise in human silhouettes is principally from three sources: the noise from errors in silhouette segmentation, noise from loose clothing and noise from occlusions. We show that in the presence of this noise a newly proposed spacing for the shape context histogram bins has the best performance.
Paper 215: Passive Error Concealment for Wavelet Coded Images with Efficient Reconstruction of High-Frequency Content
Real time image and video communication in packet switched networks suffers severely from packet loss. When using a dispersive packetization, i.e., spreading neighboring pixel data or coefficients over different packets, the lost data can be estimated from its correctly received neighbors. This paper presents a novel locally adaptive error concealment method for subband coded images. In the proposed method, we reconstruct a lost coefficient partially from its horizontal and partially from its vertical neighboring coefficients. We also estimate the optimal interpolation weights from the neighborhood of the lost coefficient. The proposed method outperforms state-of-the-art methods of similar and higher complexity. Especially, the high-frequency coefficients are reconstructed much better, resulting in better reconstruction of textures and edges. The results demonstrate a significant improvement of visual quality compared to recent related methods.
Paper 216: Fourier-Based Inspection of Free-Form Reflective Surfaces
A general free-form surface inspection approach relying on the projection of a structured light pattern and the interpretation of the generated stripe structures by means of Fourier-based features is proposed in this paper.
The major concerns of this paper are the determination of various reference sets of stripe patterns, and the detailed investigation on the subset of Fourier features that best characterizes free-form bright/dark structures. In order to tackle the inspection problem with a general approach, a first part of this paper is dedicated to the definition of different image data sets that correspond to various types of free-form specular shapes recorded with a structured illumination. A second part deals with the optimization of the most appropriate pattern recognition process. The optimization is dedicated to the use of different pattern arrangements, and the evaluation of different Fourier feature subsets.
It is shown that with only 10 Fourier features and a certain pattern arrangement, high classification rates of free-form surfaces can be obtained.
Paper 218: Graffiti detection using a Time-Of-Flight camera
Time-of-Flight (TOF) cameras relate to a very recent and growing technology which has already proved to be useful for computer vision tasks. In this paper we investigate on the use of a TOF camera to perform video-based graffiti detection, which can be thought of as a monitoring system able to detect acts of vandalism such as dirtying, etching and defacing walls and objects surfaces. Experimental results show promising capabilities of the proposed approach, with improvements expected as the technology gets more mature.
Paper 221: A Fast and Fully Automatic Ear Recognition Approach Based on 3D Local Surface Features
Sensitivity of global features to pose, illumination and scale variations encouraged researchers to use local features for object representation and recognition. Availability of 3D scanners also made the use of 3D data (which is less affected by such variations compared to its 2D counterpart) very popular in computer vision applications. In this paper, an approach is proposed for human ear recognition based on robust 3D local features. The features are constructed on distinctive locations in the 3D ear data with an approximated surface around them based on the neighborhood information. Correspondences are then established between gallery and probe features and the two data sets are aligned based on these correspondences. A minimal rectangular subset of the whole 3D ear data only containing the corresponding features is then passed to the Iterative Closest Point (ICP) algorithm for final recognition. Experiments were performed on the UND biometric database and the proposed system achieved 90, 94 and 96 percent recognition rate for rank one, two and three respectively. The approach is fully automatic, comparatively very fast and makes on assumption about the localization of the nose or the ear pit, unlike previous works on ear recognition.
Paper 222: A multiresolution robust watermarking approach for scalable wavelet image compression
This paper proposes a multiresolution blind watermarking approach in wavelet domain. The proposed approach performs a multiresolution decomposition of the logo (watermark) image. The logo insertion is started from the lowest frequency subband of the decomposed image and each decomposed logo subband is inserted into its counterpart subband of the decomposed image. The watermarked image does not show any perceptual degradation. To test the scalability features of the approach and robustness of the watermark against image compression, the watermarked image was first encoded by a highly scalable modification of SPIHT and then decoded at different bitrates and spatial resolutions. Multiple spatial resolution levels of the logo is progressively detectable from the decoded watermarked image. Experimental results confirm scalability features of the approach and its robustness against lossy compression. This approach could efficiently provide security for visual image transmission especially over heterogenous networks, where different end-users need to be differently (in quality and resolution) served according to their device capability and network access bandwidth.
Paper 224: Kernel Based Approach for High Dimensional Heterogeneous Image Features Management in CBIR Context
In this paper we address a challenge of the problem of the dimensionality curse and the semantic gap reduction for content based image retrieval in large and heterogeneous databases. The strength of our idea resides in building an effective multidimensional indexing method based on kernel principal component analysis (KPCA) which supports efficiently similarity search of the heterogeneous vectors (color, texture, shape) and maps data vectors on a low feature space that is partitioned into regions. An efficient approach to approximate feature space regions is proposed with the corresponding upper and lower distance bounds. Finally, relevance feedback mechanism is exploited to create a flexible retrieval metric in order to reduce the semantic gap between the user need and the data representation. Experimental evaluations show that the use of region approximation approach with relevance feedback can significantly improve both the quality and the CPU time of the results.
Paper 225: Video-Based Fall Detection in the Home Using Principal Component Analysis
This paper presents the design and real-time implementation of a fall-detection system, aiming at detecting fall incidents in unobserved home situations. The setup employs two fixed, uncalibrated, perpendicular cameras. The foreground region is extracted from both cameras and for each object, principal component analysis is employed to determine the direction of the main axis of the body and the ratio of the variances in x and y direction. A Gaussian multi-frame classifier helps to recognize fall events using the above two features. The robustness of the system is increased by a head-tracking module, that can reject false positives. We evaluate both performance and efficiency of the system for a variety of scenes: unoccluded situations, cases where the person carries objects and occluded situations. Experiments show that our algorithm can operate at real-time speed with more than 85% fall-detection rate.
Paper 226: Open or Closed Mouth State Detection: Static Supervised Classification Based on Log-polar Signature
The detection of the state open or closed of mouth is an important information in many applications such as hypo-vigilance analysis, face features segmentation or emotions recognition. In this work we propose a supervised classification method for mouth state detection based on retina filtering and cortex analysis inspired by the human visual system. The first stage of the method is the learning of reference signatures (Log Polar Spectrum) from some open and closed mouth images manually classified. The signatures are constructed by computing the amplitude log-polar spectrum of the retina filtered images. Principal Components Analysis (PCA) is then performed using the Log Polar Spectrum as feature vectors to reduce the number of dimension by keeping 95 % of the total variance. Finally a binary SVM classifier is trained using the projections the principal components given by the PCA in order to classify the mouth.
Paper 227: Mosaicing of Fibered Fluorescence Microscopy Video
Fibered fluorescence microscopy is a recent developed image modality using a fiber optic probe connected to a laser scanning unit. It allows for in-vivo scanning of small animal subjects by moving the probe along the tissue surface. During the scans images are continuity captured, allowing to acquire an area larger then the field of view of the probe as a video. But there is still a need to obtain a single static image from the multiple overlapping frames. In this paper we introduce a mosaicing procedure for this kind of video sequence. An additional motivation for the mosaicing is the use of overlapping redundant information to improve the signal to noise level of the acquisition, since the individual frames tend to have both high noise levels and intensity inhomogeneities.
Paper 229: Contour Detection for Industrial Image Processing by Means of Level Set Methods
We consider the problem of the automatic inspection of in- dustrial metal pieces. The purpose of the work presented in this paper is to derive a method for defect detection. For the first time in this con- text we adapt level set method to distinguish hollow regions in the metal pieces from the grinded surface. We compare this method with Canny edge enhancement and with a thresholding method based on histogram computation. The experiments performed on two industrial images show that the proposed method retrieves correctly fuzzy contours and is robust against noise.
Paper 231: MIDIAS: An Integrated 2D/3D Sensor System for Safety Applications
In this article we present an integrated micro-system consisting of a high resolution gray-value camera and a range camera. We discuss a flexible calibration method, which is essential for the three-dimensional reconstruction of the scene observed by the camera system. For the calibrated micro-system we present a simple and fast data fusion technique, which assigns distance information to each image pixel of the gray-value camera. Our methods enhance the resolution of the coarse distance information provided by the range camera. We demonstrate the applicability of our micro-system by two application examples within the safety domain: Front-view pedestrian recognition and intrusion detection with automated retrieval of the intruder image.
Paper 232: Knee Point Detection in BIC for Detecting the Number of Clusters
Bayesian Information Criterion (BIC) is a promising method for detecting the number of clusters. It is often used in model-based clustering in which a decisive first local maximum is detected as the number of clusters. In this paper, we re-formulate the BIC in partitioning based clustering algorithm, and propose a new knee point finding method based on it. Experimental results show that the proposed method detects the correct number of clusters more robustly and accurately than the original BIC and performs well in comparison to several other cluster validity indices.
Paper 233: Grey-scale Morphology with Spatially-Variant Rectangles in Linear Time
Spatially variable structuring elements outperform translation-invariant ones by their ability to locally adapt to image content.
Without restrictions, they suffer from an overwhelming computational complexity. Fast methods for their implementation have recently been proposed for 1-D functions. This paper proposes an extension to 2-D with resizable rectangles.
Paper 234: Image Segmentation Based on Supernodes and Region Size Estimation
A kind of self-adaptive image segmentation algorithm is introduced in this paper, and of which the main frame is based on Graph Structure. Two contributions have been made in our work. First, super-pixels act as the graph nodes for computational efficiency, at the same time, more local features could be abstracted from the pre-segmented image. Second, region size is estimated during the process to reduce interaction between human and computer. Experimental results demonstrate that the improved method is unsupervised and could give satisfactory segmentation.
Paper 235: Spatially-variant directional mathematical morphology operators based on a diffused average squared gradient field
This paper proposes an approach for mathematical morphology operators whose structuring element can locally adapt its orientation across the pixels of the image. The orientation at each pixel is extracted by means of a diffusion process of the average squared gradient field. The resulting vector field, the average squared gradient vector flow, extends the orientation information from the edges of the objects to the homogeneous areas of the image. The provided orientation field is then used to perform a spatially variant filtering with a linear structuring element. Results of erosion, dilation, opening and closing spatially-variant on binary images prove the validity of this theoretical sound and novel approach.
Paper 236: A Comparison of Multiclass Svm Methods for Real World Natural Scenes
Categorization of natural scene images into semantically meaningful categories is a challenging problem that requires usage of multiclass classification methods. Our objective in this work is to compare multiclass SVM classification strategies for this task. We compare the approaches where a multi-class classifier is constructed by combining several binary classifiers and the approaches that consider all classes at once. The first approach is generally termed as "divide-and-combine" and the second is known as "all-in- one". Our experimental results show that all-in-one SVM outperforms the other methods.
Paper 237: Video Denoising and Simplification via Discrete Regularization on Graphs
In this paper, we present local and nonlocal algorithms for video denoising and simplification based on discrete regularization on graphs. The main difference between video and image denoising is the temporal redundancy in video sequences. Recent works in the literature showed that motion compensation is counter-productive for video denoising. Our algorithms do not require any motion estimation. In this paper, we consider a video sequence as a volume and not as a sequence of frames. Hence, we combine the contribution of temporal and spatial redundancies in order to obtain high quality results for videos. To enhance the denoising quality, we develop a robust method that benefits from local and nonlocal regularities within the video. We propose an optimized method that is faster than the nonlocal approach, while producing equally attractive results. The experimental results show the efficiency of our algorithms in terms of both Peak Signal to Noise Ratio and subjective visual quality.
Paper 238: Configurable passband imaging spectrometer based on acousto-optic tunable filter
This work presents a new configurable imaging spectrometer called Autonomous Tunable Filtering System (ATFS). The system can be configured to acquire a single narrow spectral band, a composite multispectral image, or a broad pass-band. This flexibility is given by the use of an Acousto-Optic Tunable Filter (AOTF) driven by a programmable radio frequency (rf) signal generator. The AOTF acts as a light- diffraction element which output wavelength is selected by the frequency of an rf signal applied to it. The designed rf driver is based on a high-speed Digital-to-Analog converter, which can synthesize any composite rf waveform formed by a combination of sine signals. The images are formed through a carefully designed optical layout and acquired with a high performance digital camera. Spectral and spatial performance of the system is presented in this paper.
Paper 241: An Image Quality Metric Based on a Colour Appearance Model
Image quality metrics have been widely used in imaging systems to maintain and improve the quality of images being processed and transmitted. Due to the close relationship between image quality and human visual perception, both computer scientists and psychologists have contributed to the development of image quality metrics. In this paper, a novel image quality metric using a colour appearance model is proposed. After the physical colour stimuli of the images being compared are transformed into perceptual colour appearance attributes, the distortion measures between the corresponding attributes are used to predict the subjective scores of image quality, by use of data driven models: Multiple Linear Regression, General Regression Neural Network and Back Propagation Neural Network. Based on the data driven model used, we have developed three image quality metrics, CAM_MLR, CAM_GRNN and CAM_BPNN. The experiments show that the performance of CAM_BPNN exceeds that of two widely used image quality metrics.
Paper 242: Synchronizing Video Sequences from Temporal Epipolar Lines Analysis
This paper deals with the issue of synchronization of a multi camera system observing dynamic scenes. The developed method presented is not based on the use of local image features that are in general not robust to possible occlusions and noise. Instead, a new approach is introduced allowing a temporal alignment of video sequences using the analysis of moving object traces in scenes in the frequency or spatial domain. This method uses the stereoscopic constraint to apply a temporal correlation by analyzing epipolar lines temporal evolution. Experimental results on real data are presented, and the estimated temporal alignment are quantitatively evaluated and compared to a time truth temporal electronic device in cases of noise and occlusions.
Paper 243: Applying Open-loop Coding in Predictive Coding Systems
This paper investigates the application of open-loop coding principles in predictive coding systems. In order to cope with the drift, which is inherent in open-loop predictive coding, a novel rate-distortion (R-D) model is proposed, capturing the propagation of quantization errors in such systems. Additionally, a novel intra-frame video codec employing the transform and spatial prediction modes from H.264 is proposed. The results obtained with the proposed codec show that allocating rate based on the proposed R-D model provides gains of up to 1.9 dB compared to a straightforward rate allocation not accounting for drift. Furthermore, the proposed open-loop predictive codec provides gains of up to 2.3 dB compared to an equivalent closed-loop intra-frame video codec using the transform, prediction modes and rate-allocation from H.264. One concludes that the considered open-loop predictive coding paradigm retains the advantages of open-loop coding, and offers the possibility of further improving the compression performance in predictive coding systems.
Paper 246: Gabor Filter-Based Fingerprint Anti-Spoofing
This paper describes Gabor filter-based method to detect spoof fingerprint attacks in fingerprint biometric systems. It is based on the observation that, real and spoof fingerprints exhibit different textural characteristics. Textural measures based on Gabor energy and co-occurrence texture features are used to characterize fingerprint texture. Fingerprint image is filtered using a bank of four Gabor filters, and then a gray level co-occurrence matrix (GLCM) method is applied to filtered images to extract minute textural details. Dimensionality of the features is reduced by principal component analysis (PCA). We test features on three different classifiers: neural network, support vector machine and OneR; then we fuse all the classifiers using the "Max Rule" to form a hybrid classifier. Overall classification rates achieved with various classifiers range from ~94.12% to ~97.65%. Thus, the experimental results indicate that, the new liveness detection approach is a very promising technique, as it needs only one fingerprint and no extra hardware to detect vitality.
Paper 247: Scene Reconstruction using MRF Optimization with Image Content Adaptive Energy Functions
Multi-view scene reconstruction from multiple uncalibrated images can be solved by two stages of processing: first, a sparse reconstruction using Structure From Motion (SFM), and second, a surface reconstruction using optimization of Markov random field (MRF). This paper focuses on the second step, assuming that a set of sparse feature points have been reconstructed and the cameras have been calibrated by SFM. The multi-view surface reconstruction is formulated as an image-based multi-labeling problem solved using MRF optimization via graph cut. First, we construct a 2D triangular mesh on the reference image, based on the image segmentation results provided by an existing segmentation process. By doing this, we expect that each triangle in the mesh is well aligned with the object boundaries, and a minimum number of triangles are generated to represent the 3D surface. Second, various objective and heuristic depth cues such as the slanting cue, are combined to define the local penalty and interaction energies. Third, these local energies are adapted to the local image content, based on the results from some simple content analysis techniques. The experimental results show that the proposed method is able to well the preserve the depth discontinuity because of the image content adaptive local energies.
Paper 248: Fluid Flow Measurement in Thermographic Video Sequences by Wavelet-multiresolution Optical Flow Estimation
Variational Optical Flow estimation models have proven to be highly useful tools for both tracking (rigid) object paths and for calculating motion fields registered in digital video sequences. Specific acquisition techniques, such as infrared thermographic video, allow to carry out further studies of the fluid dynamics for several kind of phenomena. This paper presents a methodological approach to obtain a reliable estimation of the temporal evolution of thermal structures in fluid surfaces using a multiresolution scheme based on the Galerkin-Wavelet Analysis. An appropriate regularizer, adapted for the specific problem herein presented, is also introduced.
Paper 250: Breast Mass Segmentation in Mammographic Images by an Effective Region Growing Algorithm
Breast cancer is the most common cause of death among women and the most effective method for its diagnosis is mammography. However, this kind of analysis is very difficult to interpret and for this reason radiologists miss 20-30% of tumors. We propose a module for the segmentation of masses than can be implemented in a complete CADx (Computer Aided Diagnosis) system. In particular, we implement a new version of the region growing algorithm specific for this kind of images and for the constraints on computation time of this application.
Paper 251: Blur and Contrast Invariant Fast Stereo Matching
We propose a novel approach for estimating a depth-map from a pair of rectified stereo images degraded by blur and contrast change. At each location in image space, information is encoded with a new class of descriptors that are invariant to convolution with centrally symmetric PSF and to variations in contrast. The descriptors are based on local-phase quantization, they can be computed very efficiently and encoded in a limited number of bits. A simple measure for comparing two encoded templates is also introduced. Results show that, the proposed method can represent a cheap but still effective way for estimating disparity maps from degraded images, without making restrictive assumptions; these advantages make it attractive for practical applications.
Paper 253: Fuzzy Rule Iterative Feature Selection (FRIFS) with respect to the Choquet Integral apply to fabric Defect Recognition
An iterative method to select suitable features in an industrial fabric defect recognition context is proposed in this paper. It combines a global feature selection method based on the Choquet integral and a fuzzy linguistic rule classi-fier. The experimental study shows the wanted behaviour of this approach: the feature number decreases whereas the recognition rate increases. Thus, the num-ber of generated fuzzy rules is reduced.
Paper 254: Simple and Robust Optic Disc Localisation using Colour Decorrelated Templates
Automatic analysis of digital fundus images, where optic disc extraction is an essential part, is an active research topic in retinal image analysis. A simple, fast and robust optic disc localisation method using colour decorrelated templates is proposed which results an accurate location of the optic disc in colour fundus images. In the training stage, PCA is performed on the colour image data to determine an eigenspace (colour decorrelated template space) that characterises optic disc colour. The colour decorralated optic disc templates are extracted from the training images projected to the colour decorrelated template space. In the localisation stage, template matching is applied to locate optic disc using the colour decorrelated templates after projecting an unseen fundus image to the colour decorrelated template space. The proposed method is compared with a method from the literature on DIARETDB1, a publicly available fundus image database.
Paper 255: Basic Video-Surveillance with Low Computational and Power Requirements Using Long-Exposure Frames
Research in video surveillance is nowadays mainly directed towards improving reliability and gaining deeper levels of scene understanding. On the contrary, we take a different route and investigate a novel, unusual approach to a very simple surveillance task -- activity detection -- in scenarios where computational and energy resources are extremely limited, such as Camera Sensor Networks.
Our proposal is based on shooting long-exposure frames, each covering a long period of time, thus enabling the use of frame rates even one order of magnitude slower than usual -- which reduces computational costs by a comparable factor; however, as exposure time is increased, moving objects appear more and more transparent, and eventually become invisible in longer exposures. We investigate the consequent tradeoff, related algorithms and their experimental results with actual long-exposure images. Finally we discuss advantages (such as intrinsic ability to deal with low-light conditions) and disadvantages of this approach.
Paper 256: Parallel Algorithm for Concurrent Computation of Connected Component Tree
The paper proposes a new parallel connected-component-tree construction algorithm based on line independent building and progressive merging of partial 1-D trees. Two parallelization strategies were developed: the parallelism maximization strategy, which balances the workload of the processes, and the communication minimization strategy, which minimizes communication among the processes. The new algorithm is able to process any pixel data type, thanks to not using a hierarchical queue. The algorithm needs only the input and output buffers and a small stack. A speedup of 3.57 compared to the sequential algorithm was obtained on Opteron 4-core shared memory ccNUMA architecture. Performance comparison with existing state of the art is also discussed.
Paper 257: Improved Infrared Face Identification Performance using Nonuniformity Correction Techniques
In this paper, the face recognitions rate performance using infrared imagery is improved by adding nonuniformity pre-processing techniques. The infrared spectra contains the heat energy emitted by a face and it naturally present an insensitive behavior to variations in illuminations. Infrared imaging system can be formed by a Focal-Plane-Array technology, a group of photodetectors located in the focal plane of an imaging systems, but inherently present the nonuniformity as fixed-pattern noise that degrades the quality of infrared images. Additionally, this nonuniformity slowly varies over time, and depending on the technology used, this drift can take from minutes to hours. Due to this, the face identification performance is degraded over time, requiring a continuous-time calibrations method in order to maintain the face recognition rate using infrared imaging system. In synthesis, this work focuses on the evaluation of the degradation in pattern recognition performance produced by the fixed-pattern noise and the improvement when nonuniformity correction techniques is applied.
Paper 258: Active Contours without Edges and with Simple Shape Priors
In this paper, we introduce two simple shape priors into the Chan and Vese level-set method, namely, a prescribed area and a prescribed area-to-perimeter ratio of particular objects. It is remarkable that these priors may be easily incorporated into the Euler-Lagrange equation of the original method. As a side effect of our experimenting with the method, we also introduce a new probability-based level-set function, which removes the need for reinitialisation and usually, according to our experience, speeds up the computation. Finally, we also propose a method how to treat, in a simple way, the situation in which the particular objects differ in brightness. Although the mentioned changes make the segmentation more reliable, they almost do not complicate the computation. The results of experiments are also presented.
Paper 260: Distributed Smart Camera Calibration using Blinking LED
Smart camera networks are very powerful for various computer vision applications. As a preliminary step in the application, every camera in the scene needs to be calibrated. For most of the calibration algorithms, image point correspondences are needed. Therefore, easy to detect objects can be used like LEDs. Unfortunately, existing LED based calibration methods are highly sensitive to lighting conditions and only perform well in dark conditions. Therefore, in this paper, we propose a robust LED detection method for the calibration process. The main contribution to the robustness of our algorithm is the blinking behavior of the LED, enabling the use of temporal pixel information. Experiments show that accurate LED detection is already possible for a sequence length of three frames. A distributed implementation on a truly embedded smart camera is performed. Finally, a successful spatial calibration is performed with this implemented method.
Paper 261: Real-Time Hough Transform on 1-D SIMD Processors: Implementation and Architecture Exploration
In the first part of this paper, an improved slope-intercept like representation is proposed for implementation of Standard Hough Transform (SHT) on SIMD (Single-Instruction, Multiple-Data) architectures with no local indirect addressing support. The real-time implementation is realized with high accuracy on our Wireless Smart Camera (WiCa) platform. The processing time of this approach is independent of the number of edge points or the number of detected lines. In the second part, we focus on analyzing the differences between the SHT implementations on 1-D SIMD architectures with and without local indirect addressing. Three aspects are compared: total operation number, memory access/energy consumption, and memory area cost. When local indirect addressing is supported, the results show a considerable amount of reduction in total operations and energy consumption at the cost of extra chip area. The results also show that the focuses for further optimization of these two architectures are different.
Paper 262: Defocus Blur Correcting Projector-Camera System
To use a projector anytime anywhere, a lot of projector-camera based approaches have been proposed. In this paper, we focus on the focal correction technique, one of the projector-camera based approaches, reduces the effect of defocus blur which occurs when a screen is located out of projector's depth-of-field. We propose a novel method for estimating projector defocus blur on a planar screen without any special measuring images. To estimate the degree of defocus blur accurate, we extract sub-region which is well-suited for defocus blur estimation in the projection image and estimate the degree of defocus blur at each extracted region. To remove the degradation caused by the defocus blur, we pre-correct the projection image depends on the degree of defocus blur before projection. Experimental results show that our method can estimate the degree of defocus blur without projecting any special images and pre-corrected image can reduce the effect of the defocus blur.
Paper 265: Intuitionistic Fuzzy Clustering With Applications In Computer Vision
Intuitionistic fuzzy sets are generalized fuzzy sets whose elements are characterized by a membership, as well as a non-membership value. The membership value indicates the degree of belongingness, whereas the non-membership value indicates the degree of non-belongingness of an element to that set. The utility of intuitionistic fuzzy sets theory in computer vision is increasingly becoming apparent, especially as a means to coping with noise. In this paper, we investigate the issue of clustering intuitionistic fuzzy image representations. To achieve that we propose a clustering approach based on the fuzzy c-means algorithm utilizing a novel similarity metric defined over intuitionistic fuzzy sets. The performance of the proposed algorithm is evaluated for object clustering in the presence of noise and image segmentation. The results indicate that clustering intuitionistic fuzzy image representations can be more effective, noise tolerant and efficient as compared with the conventional fuzzy c-means clustering of both crisp and fuzzy image representations.
Paper 268: Fuzzy Clustering and Active Contours for Histopathology Image Segmentation and Nuclei Detection
Histopathology imaging provides high resolution multispectral images for study and diagnosis of various types of cancers. The automatic analysis of these images can greatly facilitate the diagnosis task for pathologists. A primary step in computational histology is accurate image segmentation to detect the number and spatial distribution of cell nuclei in the tissue, along with segmenting other guiding structures such as lumen and epithelial regions which together make up a gland structure. This paper presents a new method for gland structure segmentation and nuclei detection. In the first step, fuzzy c- means with spatial constraint algorithm is applied to detect the potential regions of interest, multiphase vector-based level set algorithm is then used to refine the segmentation. Finally, individual nucleus centers are detected from segmented nuclei clusters using iterative voting algorithm. The obtained results show high performances for nuclei detection compared to the human annotation.
Paper 269: Improving Image Vector Quantization with a Genetic Accelerated K-means Algorithm
In this paper, vector quantizer optimization is accomplished by a hybrid evolutionary method, which consists of a modified genetic algorithm (GA) with a local optimization module given by an accelerated version of the K-means algorithm. Simulation results regarding image compression based on VQ show that the codebooks optimized by the proposed method lead to reconstructed images with higher peak signal-to-noise ratio (PSNR) values and that the proposed method requires fewer GA generations (up to 40%) to achieve the best PSNR results produced by the conventional GA + standard K-means approach. The effect of increasing the number of iterations performed by the local optimization module within the proposed method is discussed.
Paper 271: Constrained Phase-Based Personalized Facial Feature Tracking
This work presents a technique for automatic personalized facial features localization and tracking. The approach uses a set of subgraphs corresponding to the face deformable parts which are attached to a main subgraph with nodes consisting of more stable features, some of these nodes represent the anchor points of the more deformable subgraphs.
At the node level, accurate positions are obtained by employing a Gabor phase–based disparity estimation technique. We used a modified formulation in which we have introduced a conditional disparity estimation procedure and a confidence measure as a similarity function that includes a phase difference term. A collection of trained graphs that were captured from different face deformations, are employed to correct the subgraph nodes tracking errors.
Experimental results show that the facial feature points can be tracked with sufficient precision by establishing an effective self–correcting mechanism.
Paper 273: A Robust Method For Edge-preserving Image Smoothing
Image smoothing is a critical preprocessing step in many image processing tasks. In this paper, a generalized edge-preserving smoothing model is derived from robust statistics theory, and its connections to anisotropic diffusion and bilateral filtering are established. The connections allow us to derive an improved numerical scheme in the context of a robust estimation process for edge preserving smoothing. Experiments illustrate that the proposed smoothing algorithm is capable of effectively reducing the distracting effects of noise without sacrificing image edge structures. The robust edge-preserving smoothing method is more than 3 dB better in terms of PSNR compared to anisotropic diffusion, bilateral filtering and the Bayes least squares Gaussian scale mixtures a wavelet-based method for image enhancement.
Paper 278: Object Tracking Using Naive Bayesian Classifiers
This work presents a tracking algorithm based on a set of naive Bayesian classifiers. We consider tracking as a classification problem and train online a set of classifiers which distinguish a target object from the background around it. Classifiers' voting make a soft decision about class adherence for each pixel in video frame, forming a confidence map. We use the mean shift algorithm to find the nearest peak in the confidence map, with respect to the previous position of the target. The location of that peak represents the new position of the object. The temporal adaptivity of the tracker is achieved by gradual update of a target model. The results demonstrate ability of the proposed method to perform successful tracking in different environmental conditions.
Paper 280: 3D Tracking Using Multi-View Based Particle Filters
Visual surveillance and monitoring of indoor environments using multiple cameras has become a field of great activity in computer vision. Usual 3D tracking and positioning systems rely on several independent 2D tracking modules applied over individual camera streams, fused using geometrical relationships across cameras. As 2D tracking systems suffer inherent difficulties due to point of view limitations (perceptually similar foreground and background regions causing fragmentation of moving objects, occlusions), 3D tracking based on partially erroneous 2D tracks are likely to fail when handling multiple-people interaction. To overcome this problem, this paper proposes a Bayesian framework for combining 2D low-level cues from multiple cameras directly into the 3D world through 3D Particle Filters. This method allows to estimate the probability of a certain volume being occupied by a moving object, and thus to segment and track multiple people across the monitored area. The proposed method is developed on the basis of simple, binary 2D moving region segmentation on each camera, considered as different state observations. In addition, the method is proved well suited for integrating additional 2D low-level cues to increase system robustness to occlusions: in this line, a naïve color-based (HSI) appearance model has been integrated, resulting in clear performance improvements when dealing with complex scenarios.
Paper 281: Automatic Feature-based Stabilization of Video with Intentional Motion through a Particle Filter
Video sequences acquired by a camera mounted on a hand held device or a mobile platform are affected by unwanted shakes and jitters. In this situation, the performance of video applications, such us motion segmentation and tracking, might dramatically be decreased. Several digital video stabilization approaches have been proposed to overcome this problem. However, they are mainly based on motion estimation techniques that are prone to errors, and thus affecting the stabilization performance. On the other hand, these techniques can only obtain a successfully stabilization if the intentional camera motion is smooth, since they incorrectly filter abrupt changes in the intentional motion. In this paper a novel video stabilization technique that overcomes the aforementioned problems is presented. The motion is estimated by means of a sophisticated feature-based technique that is robust to errors, which could bias the estimation. The unwanted camera motion is filtered, while the intentional motion is successfully preserved thanks to a Particle Filter framework that is able to deal with abrupt changes in the intentional motion. The obtained results confirm the effectiveness of the proposed algorithm.
Paper 283: A Generalized Appriou's Model for Evidential Classification of Multispectral Images: A Case Study of Algiers City
In this paper, we shall describe an evidential supervised classifier of multispectral satellite images. The evidence theory of Dempster-Shafer (DST) is used to take into account the ignorance and the uncertainty related to data, and so, overcome the Bayesian classifier limits. Notice that application fields of DST are initially related on multisensor, multitemporal and multiscale data fusion. In this study, our contribution lies in developing an evidential classification process that can be seen as a multisource fusion process where each predefined thematic class is considered as one source of information. The evidential mass functions of the considered thematic hypotheses are estimated using Appriou's transfer model whose we propose to generalize to a multi-class case. Developed DST-classifier has been tested on multispectral ETM+ image covering the urban north-eastern part of Algiers (Algeria). The spectral validation of obtained evidential classes allows us to confirm the accuracy of the resulting land cover map.
Paper 284: Multidimensional noise removal method based on PARAFAC decomposition
Multicomponent sensors are more and more developed since they allow to measure simultaneously several parameters. Thus, new kind of processing have been developed for some years. In this paper, we are particularly concerned with tensor signal processing for noise removal in multidimensional images. We adapt a PARAFAC based method to remove noise from multidimensional images. Some results on hyperspectral images and comparisons with a TUCKER3 based method are given.
Paper 285: Sub-Optimal Camera Selection in Practical Vision Networks through Shape Approximation
Within a camera network, the contribution of a camera to the observations of a scene depends on its viewpoint and on the scene configuration. This is a dynamic property, as the scene content is subject to change over time. An automatic selection of a subset of cameras that significantly contributes to the desired observation of a scene can be of great value for the reduction of the amount of transmitted and stored image data. We propose a greedy algorithm for camera selection in practical vision networks where the selection decision has to be taken in real time. The selection criterion is based on the information from each camera sensor's observations of persons in a scene, and only low data rate information is required to be sent over wireless channels since the image frames are first locally processed by each sensor node before transmission. Experimental results show that the performance of the proposed greedy algorithm is close to the performance of the optimal selection algorithm. In addition, we propose communication protocols for such camera networks, and through experiments, we show the proposed protocols improve latency and observation frequency without deteriorating the performance.